Research Seminaire presented by Nguyen Nhu Van, Ph.D. Students IFI - La Rochelle, at IFI, 14h00, 25 march 2010
Title : The visual representation of words and textual information retrieval by content of images
Resume : We propose an original CBIR system in which the association between textual keywords and visual features is used for image annotation and relevance feedback. In our framework, the state-of-the-art Bag of Words (BoW) model is used. A novel method to link the knowledge with visual content in which external image information (textual keywords) is represented by multiple BoWs is created. Using this method, the system is more flexible with the ability of querying by keywords and the ability of annotating images within the retrieval process. By integrating a novel cluster-based relevance feedback, the retrieval performance is better without requesting more users‘ effort. Moreover, with a long-term relevance feedback strategy, the visual representations of textual keywords are improved. Hence, the retrieval performance is better day by day and the image annotation is also refined.